CSSE463: Image Recognition Day 3

CSSE463: Image Recognition Day 3



 Announcements/reminders:

 Lab 0 should have been turned in last night.

 Tomorrow: Lab 1 (posted): on color images. Bring laptop.





 Today:

 Introduce Fruit Finder, due next Friday.

 Lots of Helpful hints in Matlab.

 Connected components and morphology





 Next week: Edge features



 Questions?

Assignment 1: Counting Fruit

 How many apples? Bananas? Oranges?

Goals

 Crash-course in using and applying Matlab

 For this reason, I will direct you to some

useful functions, but will not give details of

all of them

 Practice writing a conference-paper style

report

 Could use style similar to ICME sunset paper

 See posted rubric for expectations

Fruit-finding technique

 Observe

 What is a banana‟s “yellow” (numerically)?

 Tomorrow in lab, we‟ll see techniques

 Model

 Can you differentiate between yellow and orange? Orange and

red? (Decisions)

 Note: this isn‟t using a classifier yet; just our best guess at hand-

tuned boundaries

 Classify pixels using your model

 “Clean up” the results

 Discuss today

 Write up your results in a nice report

Region processing

 Binary image analysis

 Today, we‟ll only consider binary images

composed of foreground and background

regions

 Example: apple and non-apple

 Use find to create a mask of which pixels belong to

each

 Demo

Functions in Matlab

Contents of foo.m:



function retVal = foo(x,y,…)

…

retVal = x+y;



Can return multiple values of any type:

[mask, count, threshold] = foo(img)

Matlab How-to

 Lots of “Random” tidbits that I used in my

solution…

Neighborhoods

 Do we consider diagonals or not?



 4-neighborhood of pixel p:

 Consists of pixels from the 4 primary compass

directions from p.

 8-neighborhood of pixel p:

 Adds 4 pixels from the 4 secondary compass

directions from p.

Connected Components

 Goal: to label groups of connected pixels.

 4-connectivity vs. 8-connectivity matters

 Matlab help: search for connected

components, and use bwlabel function

 Demo

Morphological operations

 Morphology = form and structure (shape)

 For binary images

 Done via a structuring element (usually a

rectangle or circle)





 Basic operations:

 Dilation, erosion, closing, opening

Erosion

 Shrinks a region

 Removes all pixels on the boundary

 Mathematical def

 Matlab: imerode(bw, structureElt)

 structureElt (for 8 neighborhood) found by:

= strel(„square‟, 3); % for erosion using

 structureElt

3x3 neighborhood

 structureElt (for 4 neighborhood) found by:

 structureElt = strel([0 1 0; 1 1 1; 0 1 0]);

Dilation

 Enlarges a region

 Adds all background pixels adjacent to the

boundary of the foreground region.



 Matlab: imdilate(bw, structureElt)

Closing and Opening

 Closing (imclose)

 Fills internal holes in a region

 Eliminates inlets on the boundary

 Dilate, then erode

 Opening (imopen)

 Removes small regions

 Eliminates peninsulas on the boundary

 erode, then dilate



 To make dilation more aggressive,

 Dilate n times, then erode n times.

 Or, use a larger structuring element

 Example: compare dilating twice using a 3x3 square with

dilating once using a 5x5 square.